RoboCasa365

Headlines · 2

1. Oct 16
   Primotion's Paimon-0 tops several embodied-AI benchmark leaderboards

   On October 16, robot-chip startup Primotion (炎曦动力) published results for its in-house embodied foundation model Paimon-0, saying it ranks first on RoboCasa365 with a composite score of 58.1, first on VLA-Arena with a 0.73 average success rate, first overall on RoboColiseum including all four sub-leaderboards (instruction following, spatial understanding, perturbation adaptation, general manipulation), and first at the IROS Workshop Challenge with 46.599 points. The company also said it has signed intent purchase orders worth nearly 100 million yuan with several firms, covering multi-chip robot domain-control solutions and embodied-model solutions. The report describes Paimon-0 as a hierarchical asynchronous World Action Model that unifies vision, language, robot state and actions in one representation and runs reasoning and action generation asynchronously at different time scales. The leaderboard results are cited by the vendor, and the report offers no independent reproduction or third-party measurement.

   创业邦 科技 · 🔥 0
2. Oct 14
   OneWM: unified causal training lets world and action modeling reinforce each other

   An arXiv preprint submitted on 15 October (arXiv:2610.21095) by Yue Su, Cewu Lu, Xihui Liu and nine other authors introduces OneWM, a framework for unified causal training that treats actions both as conditions for world prediction and as targets for policy learning inside one shared video-action architecture. Evaluating on the LIBERO, LIBERO-Plus and RoboCasa365 embodied testbeds plus real-world policy experiments, the authors report that learning the visual consequences of specified actions improves policy performance and generalization across changing scenes, while action learning in turn improves multi-view world-prediction fidelity. The paper answers both of its motivating questions positively, concluding that world modeling and action generation benefit each other under unified causal training; all results are author-reported with no independent replication supplied.

   Hugging Face · Papers · 🔥 0

Experience and discussion from the community

Share my RoboCasa365 experienceAsk about RoboCasa365